Semiconductors are used in integrated circuits for electronic applications, including radios, televisions, cell phones, and personal computing devices, as examples. One type of semiconductor device is a semiconductor storage device, such as a dynamic random access memory (DRAM) and flash memory, which uses a charge to store information.
Various memory types are commonly used to digitally store a substantial amount of data. DRAMs have moderate cost, are very fast and can have access times on the order of tens of nanoseconds, but lose the stored data upon loss of electrical power, i.e., they are “volatile.” Present “flash” memories are non-volatile, are more expensive perhaps by a factor of ten, and have access times from tens of nanoseconds up to near a microsecond. Hard-disk drives are substantially lower in cost than DRAMs, are non-volatile, but have access times generally greater than a millisecond. Further application considerations for each technology include limitations on the number of times a memory cell can be written or read before it deteriorates, how long it reliably retains data, its data storage density, how much energy it consumes, the need for integral mechanical devices, and the complexity and expense of associated circuitry. Considering these limitations, there is now no ideal technology for general applications. Magnetic random access memory (MRAM) as described below appears to have properties that position it well for widely accepted digital memory applications, overcoming many of these limitations.
Spin electronics, which combines semiconductor technology and magnetics, is a relatively recent development in semiconductor memory devices. The spin of an electron, rather than the charge, is used to indicate the presence of a logic “1” or “0”. One such spin electronic device is a resistive memory device referred to as a magnetic random access memory, which includes conductive lines positioned perpendicular to one another in different metal layers, the conductive lines sandwiching a magnetic stack which functions as a memory cell. The place where the conductive lines intersect is called a cross-point. A current flowing through one of the conductive lines generates a magnetic field around the conductive line and orients the magnetic polarity of one layer of the magnetic stack. A current flowing through the other conductive line induces a superimposed magnetic field and can partially turn the magnetic polarity, also. Digital information, represented as a “0” or “1”, is storable in the alignment of magnetic moments in the magnetic stack. The resistance of the magnetic stack depends on the moment's alignment. The stored state is read from the magnetic stack by detecting the component's resistive state. An array of memory cells may be constructed by placing the conductive lines in a matrix structure having rows and columns, with the magnetic stack being placed at the intersection of the conductive lines.
A key advantage of MRAMs compared to traditional semiconductor memory devices, such as DRAMs, is that MRAMs are non-volatile upon removal of electrical power. This is advantageous because a personal computer (PC) utilizing MRAMs could be designed without a long “boot-up” time as with conventional PCs that utilize DRAMs, as an example.
FIG. 1 illustrates a magnetic tunnel junction (MTJ) stack that comprises a resistive or magnetic memory cell. The terms “memory cell” and “MTJ stack” are used interchangeably herein and refer to the MTJ shown in FIG. 1. The MTJ comprises two ferromagnetic layers M1 and M2 that are separated by a tunnel layer TL. The MTJ stack is positioned at the cross-point of two conductors, referred to as a wordline WL and a bitline BL. One magnetic layer M1 is referred to as a free layer, and the other magnetic layer M2 is referred to as a fixed layer. The magnetic orientation of the free layer M1 can be changed by the superposition of the magnetic fields caused by programming current IBL that is run through the bitline BL and the programming current IWL that is run through the wordline WL. A bit, e.g., a “0” or “1”, may be stored in the MTJ stack by changing the orientation of the free magnetic layer relative to the fixed magnetic layer. If both magnetic layers M1 and M2 have the same orientation, the MTJ stack has a low resistance RC. The resistance RC is higher if the magnetic layers have opposite magnetic orientations.
FIG. 2 illustrates an MRAM memory device 10 having a select transistor X1. In some MRAM memory array designs, the MTJ stack is combined with a select transistor X1, as shown in FIG. 2, which is a cross-sectional view of a 1T1MTJ design (one transistor and one MTJ stack). The 1T1MTJ design uses the select transistor X1 for selection and fast access of the MTJ during a read operation. A schematic diagram of the MTJ stack and select transistor X1 is shown in FIG. 3. As shown in FIG. 2, a bitline BL is coupled to one side of the MTJ stack, and the other side of the MTJ stack is coupled to the drain D of the select transistor X1 by metal layer MX, via VX, and a plurality of other metal and via layers, as shown. The source S of the transistor X1 is coupled to ground (GND). X1 may comprise two parallel transistors that function as one transistor, as shown in FIG. 2. Alternatively, X1 may comprise a single transistor, for example. The gate G of the transistor X1 is coupled to a read wordline (RWL), shown in phantom, that may be positioned in a different direction than, e.g., perpendicular to, the bitline BL direction.
The select transistor X1 is used to access the memory cells MTJ. In a read (RD) operation during current sensing, a constant voltage is applied at the bitline BL. The select transistor X1 is switched on, e.g., by applying a voltage to the gate G by the read wordline RWL, and current then flows through the bitline BL, the magnetic tunnel junction MTJ, over the MX layer, down the metal and via stack, through the transistor drain D, and through the transistor X1 to ground GND. This current is then measured and is used to determine the resistance of the MTJ, thus determining the programming state of the MTJ. To read another cell in the array, the transistor X1 is switched off, and the select transistor of the other cell is switched on.
The programming or write operation is accomplished by programming the MTJ at the cross-points of the bitline BL and programming line or write wordline WWL using selective programming currents. For example, a first programming current IBL passed through the bitline BL causes a first magnetic field component in the MTJ stack. A second magnetic field component is created by a second programming current IWL that is passed through the write wordline WWL, which may run in the same direction as the read wordline RWL of the memory cell, for example. The superposition of the two magnetic fields at the MTJ produced by programming currents IBL and IWL causes the MTJ stack to be programmed. To program a particular memory cell in an array, typically a programming current is run through the write wordline WWL, which creates a magnetic field at all cells along that particular write wordline WWL. Then, a current is run through one of the bitlines, and the superimposed magnetic fields switch only the MTJ stack at the cross-point of the write wordline WWL and the selected bitline BL.
The resistance difference between programmed and unprogrammed MRAM memory cells is relatively small. For example, the MTJ may be in the order of a 10 k ohm junction, and there is a change of about 30% in the resistance when a magnetic field is applied at the MTJ. This changes the sense value from 10 k ohm to between about 6 k to 8 k ohm, e.g., 7 k ohm. For other memory devices such as flash memory cells or static random access memory (SRAM) cells, there is a larger resistance difference between programmed and unprogrammed memory cells than in MRAMs. For example, if a flash cell is activated, the “on” resistance is about 5 k ohms, and the “off” resistance is substantially infinite. While other types of memory cells substantially completely switch on or off, an MRAM cell only has a small change in the resistance value upon programming. This makes MRAM cell state sensing more difficult, especially for a very rapid current sensing process that may be required for a high-speed memory.
Either current sensing or voltage sensing of MTJ resistance can be used to detect the state of memory cells. DRAMs usually are sensed using voltage sensing, for example. In voltage sensing, the bitline is precharged, e.g., to 1 volt, with the memory cell not activated. When the memory cell is activated, the memory cell charges or discharges the bitline and changes the voltage of the bitline. However, in some types of memory cells, the memory cell is small, and the bitline length may be long, e.g., may extend the entire width of the chip. The memory cell may not be able to provide enough cell current to discharge or charge a large bitline capacity within a required time. This results in an excessive amount of time being required to read the memory cells. Therefore, voltage sensing is not a preferred choice of sensing scheme for some memory devices, such as MRAM devices, because of the need to alter charge in a parasitic capacitance by a changing voltage.
Current sensing may be used to detect a resistance change of resistive memory cells. Current sensing is the desired method of sensing the state of MRAM cells, for example. In current sensing, a voltage is applied to the bitline, and the bitline voltage is kept constant with a sense amplifier. The cell current is directly measured, with the cell current being dependent on the resistance of the memory cell being read. The use of current sensing reduces the capacitive load problem from long bitlines that may occur in voltage sensing because the voltage of the sensed lines is held constant, thereby avoiding altering charge in the different interconnection capacitances of different memory cells.
In MRAM device current sensing, a constant voltage is applied to the bitline, generally as a source follower, and the current change at the bitline due to the resistance change of the magnetic tunnel junction is measured. However, because the resistance difference between a programmed and unprogrammed cell is small in MRAM memory cells, the current difference sensed is also smaller than the current change from a flash or an SRAM (static RAM) cell, for example.